A fundamental and foremost objective in biomedical research is to establish accurate, valid measurements of the clinical disease of interest. The accuracy and validity of such disease measurements are commonly established by assessing similarity between measurements made on a subject by multiple observers/time points or by comparing with some gold standard (best available) measurement. Although the foundation of the methodology for assessing accuracy of measurements has been formulated, many important and challenging statistical issues have not yet been resolved, particularly in mental health studies. These include: (1) most current methods are derived only for complete uncensored data, (2) the methods are applicable only when both measurements are made on the same scale, and (3) the methods are not widely used for constructing new instruments or developing interpretation criteria of instruments. Motivated by studies in mental health, we first propose new agreement measures for assessing agreement in the presence of incomplete or censored observations. Unlike existing measures that focus on survival times, the new indices are defined based on bivariate hazard functions or survival processes to address the special needs for handling censoring and to answer questions regarding the temporal pattern of agreement (Aim 1). In the second Aim, we propose to broaden the concept of agreement, thereby allowing for the assessment of the correspondence between instruments on different scales or of different types. We will develop innovative statistical methods to measure the broad sense agreement when (a) both scales are continuous and (b) one scale is continuous and the other is ordinal (Aim 2). Based on the methods developed in Aim 2, we will develop methods for (a) finding interpretable cut-points of a continuous scale in terms of a categorical scale and (b) developing a new psychological instrument by finding an optimal linear combination of items in an existing instrument for more accurate diagnosis of major depression among diabetes patients (Aim 3). Finally, we will disseminate our work to research communities by developing user-friendly software and creating a web site to post publications and software with user manuals (Aim 4). This proposal is designed to improve analytic methods for mental health research by developing new methodology, incorporating existing methodology and by targeting this effort toward important scientific mental health studies. These developments will directly benefit mental health research, but they are ubiquitous enough to be generally useful contributions to statistical practice. PUBLIC HEALTH RELEVANCE: The proposed research project will develop new analytical methods for improving the accuracy of measurements in mental health studies. We develop methods for creating new instruments that are easy to use for screening depression. We also develop analytic methods to determine whether expensive, time consuming instrument can be replaced by inexpensive, easy-to-use instrument for diagnosing depression.